Deep learning-based evaluation of prostate lesions useful during active surveillance
Click Here to Manage Email Alerts
A deep-learning algorithm detected prostate lesions on multiparametric MRI and tracked lesions throughout the disease course among a subset of patients on active surveillance for prostate cancer, according to study results.
The findings — presented at American Urological Association Annual Meeting — also showed patients with aggressive features that the deep-learning system predicted had a higher likelihood of progression off active surveillance.
Rationale and methods
“[Artificial intelligence] and deep-learning algorithms are currently a main area of focus, especially in prostate cancer,” Michael Daneshvar, MD, urologic oncology fellow at NCI/NIH, told Healio. “We therefore aimed to find specific areas within prostate cancer where we could impact lesion detection and disease follow-up, as active surveillance is currently an area where multiparametric MRI has not been established.”
Researchers used a deep-learning detection and grading system to automatically evaluate lesion dynamics on multiparametric MRI among 49 men undergoing active surveillance for prostate cancer. They calculated total lesion volume growth and year and used Cox proportional hazard regression analysis to assess the association of deep learning-based output with disease progression.
Men underwent a median four scans (range, 2-6) during a median 5 years (range, 8.1 months to 8.2 years).
Key findings
Results showed no significant differences between men who progressed (n = 24) and those who did not progress (n = 25) for deep learning-based lesion, lesion burden or maximum risk scores at baseline.
However, compared with nonprogressors, researchers found a higher number of men who experienced progression had high-risk lesions (n = 21 vs. 15; P = .05). Progressors also had a higher burden of disease (0.022 vs. 0.007) and experienced a median 0.17 cc/year tumor growth rate compared with 0.04 cc/year among those without disease progression.
Men who harbored persistent high-risk lesions on follow-up imaging had higher risk for progression (HR = 3.3; 95% CI, 0.98-11).
“Although we are not at a point where AI can substitute biopsy, this algorithm can certainly help augment our decision-making pathways,” Daneshvar said.
Implications
“Specific number of lesions and burden of lesions within the prostate can be tracked throughout time, which can help identify areas of progression that can either lead to continued surveillance or biopsy,” Daneshvar said. “Because this was a pilot study, we would certainly hope to expand the scope of the study and cohort and continue tracking this cohort. Also, a prospective, multi-institutional cohort would be essential in the future.”
Collapse
Daneshvar M, et al. Abstract PD40-05. Presented at: American Urological Association Annual Meeting; May 13-16, 2022; New Orleans.
Read more about
Click Here to Manage Email Alerts